Summary of Rasa: Building Expressive Speech Synthesis Systems For Indian Languages in Low-resource Settings, by Praveen Srinivasa Varadhan et al.
Rasa: Building Expressive Speech Synthesis Systems for Indian Languages in Low-resource Settings
by Praveen Srinivasa Varadhan, Ashwin Sankar, Giri Raju, Mitesh M. Khapra
First submitted to arxiv on: 19 Jul 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Machine Learning (cs.LG); Sound (cs.SD); Audio and Speech Processing (eess.AS)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Rasa is a multilingual expressive text-to-speech (TTS) dataset for Indian languages, comprising 10 hours of neutral speech and 1-3 hours of expressive speech for each of the 6 Ekman emotions in Assamese, Bengali, and Tamil. The dataset is evaluated using MUSHRA scores, revealing that just 1 hour of neutral data and 30 minutes of expressive data can achieve a Fair system. Increasing neutral data to 10 hours with minimal expressive data significantly enhances expressiveness, providing a practical recipe for resource-constrained languages. The study highlights the importance of syllabically balanced data and pooling emotions to improve expressiveness. Additionally, challenges in generating specific emotions like fear and surprise are discussed. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary We created a new way to make text-to-speech (TTS) systems work for Indian languages. Our dataset, called Rasa, has lots of different sounds and emotions, including happy, sad, scared, and surprised faces. We tested our system and found that even with just a little bit of data, it can still be pretty good. But if we add more neutral speech (like talking about everyday things) and some expressive speech (like acting out emotions), the system gets even better. This is helpful because not all languages have lots of data available, so this way we can make them work too. |
